++
In the 1950s Herbert Jasper and colleagues pioneered chronic microelectrode recordings from alert animals engaged in natural behaviors. This approach, which allows researchers to study the activity of single neurons while animals perform a controlled behavioral task, has made enormous contributions to our knowledge of the neuronal mechanisms underlying many brain functions. A microelectrode can also be used to deliver weak electrical currents to a small volume of tissue around its tip. When used in the cerebral cortex, this technique is called intracortical microstimulation.
++
These methods have been complemented more recently by techniques that can be used in human subjects, such as functional imaging and transcranial magnetic stimulation. Nearly every insight that will be described in the rest of this chapter and in Chapter 38 has been derived from these techniques.
++
Edward Evarts, the first to use chronic microelectrode recordings to study the primary motor cortex in behaving monkeys, made several discoveries of fundamental importance. He found that single neurons in this area discharge during movements of a limited part of the contralateral body, such as one or two adjacent joints in the hand, arm, or leg (Figure 37–9). Some neurons discharge during flexion of a particular joint and are reciprocally suppressed during extension, whereas other cells display the opposite pattern. This movement-related activity typically begins 50 to 150 ms before the onset of agonist muscle activity. These pioneering studies suggested that single neurons in primary motor cortex generate signals that provide specific information about movements of specific parts of the body before those movements are executed.
++
++
Many subsequent studies have provided further insight into the contribution of different cortical motor areas to the control of voluntary movements. In general, the output signals from premotor areas are strongly dependent on the context in which the action is performed, such as the stimulus-response associations and the rules that guide which movement to make. In contrast, the commands generated by the primary motor cortex are more closely related to the mechanical details of the movement and are usually less influenced by the behavioral context. However, the relative role of these different areas to voluntary motor control, including the primary motor cortex itself, continues to be an area of active research and controversy. The rest of this chapter and Chapter 38 describe our current understanding of the different roles of cortical motor areas.
++
Columnar arrays of neurons with similar response properties are a prominent feature of many sensory areas of cortex. It is surprising therefore that there is only weak evidence for such functional columns in the primary motor cortex. The cell bodies and apical dendrites of primary motor cortex neurons tend to form radially oriented columns. The terminal arbors of thalamocortical and corticocortical axons form localized columns or bands and corticomotoneurons tend to cluster in small groups with similar muscle fields. Motor cortex neurons recorded successively as a microelectrode descends perpendicularly through the neuronal layers between the pial surface and the white matter typically discharge during movements of the same part of body and can have similar preferred movement directions. Nevertheless, adjacent cells often show very different response patterns.
+++
Motor Commands Are Population Codes
++
The complex overlapping organization of the motor map for the arm and hand suggests at least two different ways to generate the motor command for a given movement. The map could function as a look-up table within which a desired movement is generated by selective activation of a few sites whose combined output produces all the required muscle activity and joint motions. Or it could be a distributed functional map in which many sites contribute to each motor command.
++
Apostolos Georgopoulos and colleagues recorded from the primary motor cortex while a monkey reached in different directions from a central starting position toward targets arrayed on a circle in the horizontal plane. Individual neurons responded during many movements, not just a single one (Figure 37–10A). Each neuron's activity was strongest for a preferred direction and often weakest for the opposite direction, as Evarts had found for single-joint movements. However, each cell also responded in a graded fashion to directions of movement between the preferred and the opposite directions. Its activity pattern thus formed a broad directional tuning curve, maximal at the preferred direction and decreasing gradually with increasing difference between the preferred direction and the target direction.
++
++
Different cells had different preferred directions, and their tuning curves overlapped extensively. All directions were represented in the neuronal population. Cells with similar preferred directions were located at several different sites in the arm motor map, and nearby cells often had different preferred directions. As a result, many cells with a broad range of preferred directions discharged at different intensities at many locations across the arm motor map during each reaching movement.
++
Despite the apparent complexity of the response properties of single neurons, Georgopoulos found that the global pattern of activity of the entire population provided a clear signal for each movement. He represented each cell's activity by a vector pointing in the cell's preferred direction. The vector's length for each direction of movement was proportional to the mean level of activity of that cell averaged over the duration of the movement (Figure 37–10B). This vectorial representation implied that an increase of activity of a given cell is a signal that the arm should move in the cell's preferred direction, and that the strength of this directional influence varies continuously for different reach directions as a function of the neuron's directional tuning.
++
Vectorial addition of all of the single-cell contributions to each output command produces a population vector that corresponds closely to the actual movement direction. That is, an unambiguous signal about the desired motor output is encoded by the summed activity of a large population of active neurons throughout the arm motor map in the primary motor cortex. As a result, neurons in all parts of the arm motor map contribute to the motor command for each reaching movement, and the pattern of activity across the motor map changes continuously as a function of the intended direction of the reaching movement.
++
Andrew Schwartz and colleagues used the same population-vector analysis to represent temporal variations in the activity of populations of primary motor cortex neurons every 25 ms while monkeys performed continuous arm movements. In the resulting time sequence of population vectors, each vector predicts the instantaneous direction and speed of the motion of the monkey's arm approximately 100 ms later (Figure 37–11). These results show that the pattern of neural activity distributed across the arm motor map varies continuously in time during complex arm movements, signaling the moment-to-moment details of the desired movement.
++
++
Further studies have confirmed that similar population-coding mechanisms are used in all cortical motor areas. This common coding mechanism undoubtedly facilitates the communication of movement-related information between the multiple areas of motor cortex during voluntary behavior.
+++
The Motor Cortex Encodes Both the Kinematics and Kinetics of Movement
++
Population-vector analyses show that neural activity in the primary motor cortex contains information about the trajectory of hand motions during reaching and drawing movements. However, to execute those movements the motor system must implement the desired motions by generating particular patterns of muscle activity.
++
Electrical stimulation of the primary motor cortex readily evokes muscle contractions, and some cells in this region have direct access to spinal motor neurons. Indeed, it was long assumed that the major role of the primary motor cortex was to specify the muscle activity that generates voluntary movements. Because muscle contractions generate the forces that displace a joint or limb in a particular direction, a critical question is whether primary motor cortex neurons signal the desired spatiotemporal form of a behavior or the forces and muscle activity required to generate the movement. That is, do these neurons encode the kinematics or the kinetics of an intended movement (Box 37–2)?
++
Box 37–2 The Equilibrium-Point Hypothesis of Movement
Most theoretical and neurophysiological studies of neural control of movement are based on variants of the force-control hypothesis, which states that the motor system controls a movement by planning and controlling its causal dynamic forces or muscle activity.
The position-control or equilibrium-point hypothesis, however, argues that cortical motor centers do not compute inverse kinematics or dynamics to specify the necessary muscle activity. Instead, this model proposes that the output from the motor cortex signals the desired spatial endpoints and equilibrium configurations of the arm and body, that is, the posture in which all external and internal (muscular) forces are at balance and no further movement occurs.
According to the equilibrium-point hypothesis, the motor cortex causes a movement of part or all of the body by generating a signal specifying a particular equilibrium or referent configuration. This descending signal exploits spinal reflex circuits and the spring-like biomechanical properties of muscles to change muscle activity and create an imbalance between external and internal forces, causing the limb to move until equilibrium is restored.
If no external force is applied to the limb, the desired, signaled, and actual equilibrium configurations should all correspond. If the motor system is confronted with an external force, however, it must signal a different referent configuration whose internal forces compensate for the external forces. Thus, according to the equilibrium-point hypothesis, the motor cortex commands the desired movement without computing the complex transformations required to encode the required forces and muscle activities. According to the hypothesis, the inverse-kinematics and inverse-dynamics transformations occur implicitly at the local spinal cord circuits and in the motor periphery itself.
++
Kinematics refers to the parameters that describe the spatiotemporal form of movement, such as direction, amplitude, speed, and path. Kinetics concerns the causal forces and muscle activity. It is also useful to distinguish the dynamic forces that cause movements from the static forces required to maintain a given posture against constant external forces such as gravity.
++
Evarts was the first to address this question with single-neuron recordings. Using a system of pulleys and weights, he applied a load to the wrist of a monkey to pull the wrist in the direction of flexion or extension. To make a particular movement the animal had to alter its level of muscle activity to compensate for the load. As a result, the kinematics (direction and amplitude) of wrist movements remained constant but the kinetics (forces and muscle activity) changed with the load. The activity of many primary motor cortex neurons associated with movements of the hand and wrist increased during movements in their preferred direction when the load opposed that movement but decreased when the load assisted it (Figure 37–12). These changes in neural activity paralleled the changes in muscle activity required to compensate for the external loads. This was the first study to show that the activity of many primary motor cortex neurons is more closely related to how a movement is performed, the kinetics of motion, than to what movement is performed, the corresponding kinematics.
++
++
A later study confirmed this property of motor cortex activity during whole-arm reaching movements. A monkey made arm movements exactly as in the task used by Georgopoulos (Figure 37–10), but additional external loads pulled the arm in different directions. To continue to move the arm along the same path, the monkey had to change the activity of its arm muscles to counteract the external loads. The level of activity of many motor cortex neurons changed systematically with the direction of the external load even though the movement path did not change. When the load opposed the direction of reach, the single-cell and total population activity increased. When the load assisted the reaching direction, the neural activity decreased in a manner that signaled the change in muscle activity and output forces required to make the movement (Figure 37–13).
++
++
Other studies have examined the issue whether the primary motor cortex organizes the kinematics or kinetics of movement by using tasks in which subjects generate isometric forces against immovable objects rather than moving the arm. The activity of many primary motor cortex neurons varies with the direction and level of static isometric output forces generated across a single joint, such as the wrist or elbow, as well as during precise pinches with the thumb and index finger (Figure 37–14A). At least over part of the tested range these responses vary linearly with the level of static force. When a monkey uses its whole arm to exert isometric force in different directions, the activity of many motor cortex neurons varies systematically with force direction, and the directional tuning curves resemble those for activity associated with reaching movements (Figure 37–14B). Because no movement is intended or produced in isometric tasks, this strongly suggests that the primary motor cortex contributes to the control of static and dynamic output forces during many motor actions.
++
++
Finally, several studies have found that the activity of some motor cortex neurons can be correlated with the detailed contraction patterns of specific muscles during such diverse tasks as isometric force generation, precision pinching of objects between the thumb and index finger, and complex reaching and grasping actions (Figure 37–15).
++
++
These findings show that some neurons in the primary motor cortex can provide information about the causal forces and muscle activity of motor outputs. Nevertheless, the activity of other neurons in the primary motor cortex appears to signal the desired kinematics of arm and hand movements rather than their kinetics, or the desired direction of isometric force but not its magnitude. Perhaps most surprisingly, the activity of some corticomotoneurons does not always correlate with the contraction of their target muscles. For instance, some corticomotoneurons discharge strongly while a monkey generates weak contractions of the target muscles to make carefully controlled delicate movements of the hand and fingers, but are nearly silent when the monkey generates powerful contractions of the same muscles to make brisk, forceful movements.
++
How can we reconcile these apparently contradictory findings about the role of the primary motor cortex in the control of movement? According to the serial model of motor control all of the neurons in the primary motor cortex should have similar properties and so should represent either the kinematics or kinetics of the desired movement, but not both. However, the experimental evidence suggests that a strictly serial model is too simplistic. The response properties of primary motor cortex neurons are not homogeneous. Signals about both the desired kinematics and required kinetics of movements may be generated simultaneously in different, or possibly even overlapping, populations of primary motor cortex neurons. Rather than representing only what movement to make (kinematics) or how to make it (kinetics), the true role of the motor cortex may be to perform the transformation between these two representations of voluntary movements.
++
Delineating the movement-related information encoded in motor cortex activity is increasingly important for the development of brain-controlled interfaces and neuroprosthetic controllers that allow patients with severe motor deficits to control remote devices such as a computer cursor, a wheelchair, or a robotic limb by neural activity alone (Box 37–3).
++
Box 37–3 Enhancing the Quality of Life of Neurological Patients: Brain-Machine Interfaces
Every year thousands of people suffer severe spinal cord trauma, subcortical strokes, or degenerative neuromuscular diseases such as multiple sclerosis and amyotrophic lateral sclerosis. Although their cortical motor systems remain largely intact and they try strenuously to move, they cannot convert their willful intentions into physical action.
These patients must depend on caregivers to attend to even their most basic needs. One of the greatest quality-of-life issues for these patients is the loss of autonomy resulting from the inability to move and sometimes even to communicate. Several technological solutions have been sought to enhance the autonomy of such patients.
One approach has been to use electroencephalographic activity recorded by scalp electrodes as a control signal for remote devices such as computer cursors or robotic tools. An alternative approach has been to record the eye movements of subjects and to use them as the control signals.
However, both methods have significant limitations. Electroencephalographic control often takes months to master because the subjects must learn how to synchronize the activity of large populations of neurons within a cortical region to generate an electrical signal that is recordable and discriminable in real time and without extensive averaging of multiple repetitions.
Eye-movement methods are much easier to implement and learn, but they prevent subjects from looking toward other objects of interest while attempting to perform a task. Moreover, both approaches require intense concentration and the focused attention of the subjects to the virtual exclusion of all other activities.
A major recent advance has been the development of brain-machine interfaces (also often called brain-computer, brain-controlled, or neuroprosthetic interfaces). This technology records neural activity reflecting the motor intentions of the individual and converts this activity into control signals for external devices. It exploits the discovery that information about static arm postures and the direction and velocity of arm movements can be extracted from the activity patterns of neuronal populations in the primary motor cortex and other arm movement-related areas of the cerebral cortex.
Brain-machine interfaces include four basic components:
Implantable electrode arrays and associated hardware to record the activity of neuronal populations in a cortical area.
Computer algorithms to extract signals about the motor intentions of the individual.
Interfaces to convert the extracted signals into control signals to generate the desired action by an external effector.
Sensory feedback signals to improve performance.
Originally tested in experimental animals, brain-machine interfaces are now undergoing clinical trials in human neurological patients. Severely paralyzed patients with multi-electrode arrays in the primary motor cortex are quickly able to learn to control a cursor on a computer monitor so as to operate computer programs, compose messages, track the random motions of a moving target, and control a simple robotic arm. The subjects are able to control the remote effectors merely by thinking about making the corresponding movements.
The centrally generated intentions activate motor cortex neurons in a manner similar to that during normal movements. The subjects can control the devices while at the same time engaged in other activities such as looking around the laboratory or even engaging in conversations. This ability dramatically illustrates the fact that much of the cortical activity that converts a motor intention into overt action occurs in the subconscious.
The initial studies using this technology demonstrated that electrodes implanted in different cortical areas yield different types of neural signals. Electrodes in the primary motor cortex provide the best signals for continuous control of the time-varying details of the kinematics and kinetics of the trajectory of a robotic device. Such control is particularly useful for tasks like manipulating objects and for making complex movements as in drawing or writing.
In contrast, signals from the premotor cortex and posterior parietal cortex may be more appropriate for specifying the overall goals and desired outcome of an action, such as the final target location, without elaborating the details of how to accomplish the goal.
A brain-controlled interface that uses a combination of signals from different cortical areas might afford a level of context-dependent control that resembles the normal voluntary control of behavior.
+++
Hand and Finger Movements Are Directly Controlled by the Motor Cortex
++
The monosynaptic projection from the primary motor cortex onto spinal motor neurons is most dense for muscles of the distal arm, hand, and fingers. This arrangement allows the primary motor cortex to regulate the activity of those muscles directly, in contrast to its indirect regulation of muscles through the reflex and pattern-generating functions of the spinal circuits. It also provides primates and humans with a greatly enhanced capacity for individuated control of hand and finger movements. Large lesions of the primary motor cortex permanently destroy this capacity.
++
Although monkeys and humans can make isolated movements of the thumb and fingers, most hand and finger actions involve combinations of stereotypical hand and finger configurations and coordinated wrist and digit movements. This has led to the hypothesis that separate cortical circuits selectively control these different stereotypical hand actions, and that the primary motor cortex converts these signals into more specific motor commands (see Chapter 38).
++
The anatomy of the muscles of the wrist and fingers further complicates the commands for individuated finger and hand movements. Several muscles have long, bifurcating tendons that act across several joints and even act on several fingers rather than just one. As a consequence, individuated control of hand and finger movements requires highly specific patterns of activation and inhibition of multiple muscles.
++
Cortical neurons controlling the hand and digits occupy the large central core of the primary motor cortex motor map but also overlap extensively with populations of neurons controlling more proximal parts of the arm (see Figure 37–2A). Some neurons within the central core discharge preferentially during movements of a single digit, but many discharge during coordinated movements of several digits, and even of the wrist and more proximal joints. Neurons that discharge during movements of different digits are distributed throughout the motor map in an extensively overlapping fashion. As a result, neural activity required to generate an individuated action of the hand and digits is distributed broadly across the distal arm and hand areas of the motor map, as is also the case for the output to more proximal parts of the arm.
++
This highly intermixed organization of the hand and digit motor map stands in striking contrast to the much more highly ordered representation of tactile sensory inputs from different parts of the hand and digits in the primary somatosensory cortex. This difference likely reflects differences in the cortical mechanisms required to analyze the spatiotemporal distribution of tactile input on the hand and digits versus those needed to coordinate individuated movements of the digits and hand.
+++
Sensory Inputs from Somatic Mechanoreceptors Have Feedback, Feed-Forward, and Adaptive Learning Roles
++
Many primary motor cortex neurons receive sensory input from proprioceptors or cutaneous mechanoreceptors. The tactile input is particularly prominent on neurons implicated in the control of hand and digit movements. These inputs inform the motor system about the current state of the body, such as the position, posture, and movement of the arm and hand and their interactions with the environment. This information can play at least three functional roles: in feedback control of ongoing movements, in feed-forward control of intended movements, and as a teaching signal during motor learning.
++
Sensory feedback from the arm provides information about both the progress of an ongoing arm movement and deviations from the intended path that should be corrected. Feedback corrections during movement are implemented by neural circuits at many levels of the motor system, ranging from reflex responses in the spinal cord to corrective adjustments of voluntary motor commands from the motor cortex. Similarly, the activity of many neurons in the primary motor cortex that control hand movements is strongly influenced by tactile stimuli on the glabrous surface of the digits and palm of the hand. This tactile input helps adjust the output signal from hand-related neurons to ensure that the subject applies enough force to the surface of an object to grasp and manipulate it, but not to crush it or let it slip.
++
Sensory feed-forward control involves continuously adjusting the level and distribution of neuronal activity throughout the cortical motor map to reflect the limb's current state of posture and movement. By pretuning the pattern of activity in the motor-cortical map and spinal motor apparatus as a function of the limb's motor state before the onset of a movement, somatic sensory input helps to assure that the appropriate motor command is generated in the motor cortex and converted into the appropriate patterns of muscle activity at the spinal level.
++
Finally, sensory input can provide information about errors experienced during movement that could be used by adaptive motor circuits to make changes to future motor commands, thus facilitating motor learning.
+++
The Motor Map Is Dynamic and Adaptable
++
The mediolateral sequence of major body segments in the motor map is highly consistent across individuals, but the details in each functional subregion can vary. This suggests that the motor map is continually shaped by an individual's motor experience. The dynamic nature of the map has been demonstrated in several ways. For instance, functional reorganization often occurs after a focal lesion so that some of the movements that had been evoked by the injured tissue are now generated by the adjacent cortex. This reorganization likely contributes to the recovery of function after local infarcts.
++
Learning a motor skill can also induce reorganization. Randy Nudo and colleagues trained monkeys to use precise movements of the thumb, index finger, and wrist to extract treats from a small well. After a monkey had become adept at the task, the area of its motor map in which intracortical microstimulation could evoke the skilled movements was larger than before training (Figure 37–16). If the monkey did not practice the task for a lengthy period, its skill level decreased, as did the cortical area from which the relevant movements could be elicited. Similar modifications of the representation of practiced actions have also been demonstrated in human motor cortex by functional imaging and transcranial magnetic stimulation.
++
++
John Donoghue and colleagues demonstrated that these adaptive changes depend on horizontal connections and local inhibitory circuits. They found two adjacent sites in the rat's motor map at which intracortical microstimulation caused contractions of muscles in the upper lips or forearm (Figure 37–17A). Within minutes after transection of the facial nerve innervating the lip muscles, stimulation of the lip-muscle site began to evoke contractions of forearm muscles.
++
++
In a related experiment they injected bicuculline into a forearm-muscle site in the motor cortex of an intact rat without a facial nerve transection to block the neurotransmitter GABA (γ-aminobutyric acid). Within minutes stimulation of the lip-muscle site evoked contractions of both lip and forearm muscles. They concluded that stimulation of the lip-muscle site activated local horizontal axons that projected into the forearm-muscle site, activity that was normally suppressed by GABAergic inhibitory interneurons (Figure 37–17B).
+++
The Motor Cortex Contributes to Motor Skill Learning
++
One of the most remarkable properties of the brain is the adaptability of its circuitry to changes in the environment—the capacity to learn from experience and store the acquired knowledge as memories. When human subjects practice a motor skill their performance improves. Important advances have been made in understanding the mechanisms underlying the learning of motor skills, also known as procedural learning (see Chapter 66). For instance, Donoghue and colleagues found an increase in the synaptic strength of local horizontal connections between different parts of the arm motor map in rats that became increasingly skilled at reaching through a small hole in a transparent barrier to grasp, retrieve, and eat small food pellets.
++
Adaptation to perturbations of movement caused by external forces has been studied extensively in human subjects. One type of force field pushes on the arm in a direction perpendicular to the direction of the arm's movement; the strength of this force increases with movement speed. Although such viscous curl fields may seem odd, they are exactly the kind of forces that act on an arm when a person reaches out while simultaneously turning his or her body. Normally, these coriolis forces do not deflect the arm movement from its intended path because your motor system has learned to predict that these forces will arise during this natural behavior and generates a motor command that corrects for them in advance.
++
However, when a subject is stationary and unexpectedly encounters an experimentally generated viscous curl field for the first time during an arm movement, the arm is deflected sideways from its usual, nearly straight path and the hand path becomes curved. When the subject makes repeated movements in the same field, the movement paths become incrementally straighter until they are indistinguishable from movements without the curl field. If the force field is then unexpectedly turned off, the path of movement curves strongly in the opposite direction (Figure 37–18A). This after-effect demonstrates that the subject has changed the motor command required to produce the desired straight movement in anticipation of the perturbing effect of the force field.
++
++
As a subject adapts to the force field, motor behavior changes from feedback correction for actual perturbations to predictive feed-forward compensation for expected perturbation. Motor-learning theory suggests that this adaptive process may involve at least two distinct learning mechanisms, known as feedback-error learning and supervised learning.
++
In feedback-error learning sensory signals about the experienced error both guide the correction for the immediate perturbation and alter adaptive feedback control circuits to permit more efficient compensation for expected perturbation. In supervised learning the motor system gradually adapts internal models, neural circuits that learn the relationship between desired movements and required motor commands in that environment (see Chapter 33). An internal forward model estimates the state of the limb in the near future based on an efference copy of the motor command and sensory feedback of the ongoing movement, and uses this estimate to generate an error signal proportional to the deviation of the estimated movement from its desired kinematics. An internal inverse model uses this and other error signals to learn how to generate the motor command that will produce a desired movement by compensating in a predictive manner for the anticipated perturbation. Neural circuits that constitute these internal forward and inverse models are thought to be located in several brain structures, including the cerebellum, superior parietal cortex, premotor cortex, and primary motor cortex.
++
Emilio Bizzi and colleagues recorded the activity of the same primary motor cortex neurons over several hours in monkeys as the animals first made arm movements without an external force field, then while they made many movements to adapt to a viscous curl field, and finally while they readapted to the baseline condition (the "washout" period). As the monkeys adapted to the force field the directional tuning of many neurons gradually changed by 15 to 20 degrees from what it was before exposure to the viscous curl field, and then rotated back to the baseline during the washout period (Figure 37–18B). Arm muscles showed similar changes during adaptation and washout, implicating those neurons in the incremental adaptation of the motor command to the external curl field. Other neurons did not change directionality during either adaptation or washout, as if their signals communicated the desired movement kinematics across all force-field conditions.
++
Two other groups of neurons showed special properties. The directional tuning of one group changed when the monkeys switched from the null field to the curl field but did not return to baseline during washout (Figure 37–18B). The other group did not change during the original adaptation from null field to curl field but changed during washout. Bizzi proposed that these two groups of neurons retain the memory of one or the other of the successive learning episodes—adaptation and washout—in the task. That is, even though the motor performance of the monkeys returned to baseline, the functional state of the primary motor cortex did not revert to its original condition—a trace of the recent learning history persisted in the altered tuning properties of some neurons.
++
These and similar findings from other studies suggest that the motor map of the primary motor cortex is not static. Instead, the neuronal circuitry creates a dynamic, adaptive map that generates the motor commands required to accomplish desired actions under different conditions. This strongly implicates the primary motor cortex in the acquisition, retention, and recall of procedural skills, but does not clarify whether it functions primarily as part of a feedback controller, as an inverse internal model for task dynamics, or both.
++
Furthermore, recent studies have found that adaptive changes in motor cortex activity lag the improvement in motor performance by several trials during adaptation. This suggests that learning-related adjustments to motor commands are initially made elsewhere, with the cerebellum as one strong candidate. The primary motor cortex may thus be more strongly involved in the slower processes of long-term retention and recall of motor skills rather than the initial phase of learning a new skill.